Title: Throughput Optimal Decentralized Scheduling with Single-bit State Feedback for a Class of Queueing Systems

Abstract: Motivated by medium access control for resource-challenged wireless Internet of Things (IoT), we consider the problem of queue scheduling with reduced queue state information. In particular, we consider a time-slotted scheduling model with $N$ sensor nodes, with pair-wise dependence, such that Nodes $i$ and $i + 1,~0 < i < N$ cannot transmit together. We develop new throughput-optimal scheduling policies requiring only the empty-nonempty state of each queue that we term Queue Nonemptiness-Based (QNB) policies. We propose a Policy Splicing technique to combine scheduling policies for small networks in order to construct throughput-optimal policies for larger networks, some of which also aim for low delay. For $N = 3,$ there exists a sum-queue length optimal QNB scheduling policy. We show, however, that for $N > 4,$ there is no QNB policy that is sum-queue length optimal over all arrival rate vectors in the capacity region.
We then extend our results to a more general class of interference constraints that we call cluster-of-cliques (CoC) conflict graphs. We consider two types of CoC networks, namely, Linear Arrays of Cliques (LAoC) and Star-of-Cliques (SoC) networks. We develop QNB policies for these classes of networks, study their stability and delay properties, and propose and analyze techniques to reduce the amount of state information to be disseminated across the network for scheduling. In the SoC setting, we propose a throughput-optimal policy that only uses information that nodes in the network can glean by sensing activity (or lack thereof) on the channel. Our throughput-optimality results rely on two new arguments: a Lyapunov drift lemma specially adapted to policies that are queue length-agnostic, and a priority queueing analysis for showing strong stability.